feat: add cache TTL configurations for kubelet stats and Linux agent

charts/cmk-rustik/templates/metrics-cache/deployment.yaml

@@ -53,6 +53,8 @@ spec:
             - --writer-allowlist={{ .Release.Namespace }}:{{ include "rustik.fullname" . }}-metrics-fetcher
             - --cluster-name={{ required "clusterName is required (there is no safe default: it names every generated host)" .Values.clusterName }}
             - --cluster-host-name={{ required "clusterHostName is required (it must exactly match the Checkmk host representing this cluster)" .Values.clusterHostName }}
+            - --kubelet-stats-cache-ttl={{ .Values.metricsCache.kubeletStatsCacheTtl }}
+            - --linux-agent-cache-ttl={{ .Values.metricsCache.linuxAgentCacheTtl }}
             {{- if .Values.hostLabels.importAllAnnotations }}
             {{- if .Values.hostLabels.importKeyPattern }}
             {{ fail "hostLabels.importKeyPattern should not be set while hostLabels.importAllAnnotations is true" }}

charts/cmk-rustik/values.yaml

@@ -95,6 +95,14 @@ metricsCache:
   # Inherits top-level logLevel by default.
   logLevel: ""
 
+  # kubeletStatsCacheTtl is how long (seconds) kubelet stats entries are
+  # persisted in the cache before expiring.
+  kubeletStatsCacheTtl: 120
+
+  # linuxAgentCacheTtl is how long (seconds) Linux agent entries are
+  # persisted in the cache before expiring.
+  linuxAgentCacheTtl: 120
+
   # extraVolumes is volume definitions to add to the Pod in the Deployment.
   extraVolumes: []
   # extraVolumeMounts is volume mounts to mount in the metrics-cache container.

metrics-cache/src/cli_args.rs

@@ -78,14 +78,21 @@ pub struct CliArgs {
     )]
     pub writer_allowlist: Vec<String>,
 
-    /// How long (seconds) entries are persisted in the cache
+    /// How long (seconds) kubelet stats entries are persisted in the cache
     #[arg(
-        short = 't',
-        long = "cache-ttl",
+        long = "kubelet-stats-cache-ttl",
         value_parser = parse_duration_secs,
         default_value = "120"
     )]
-    pub cache_ttl: Duration,
+    pub kubelet_stats_cache_ttl: Duration,
+
+    /// How long (seconds) Linux agent entries are persisted in the cache
+    #[arg(
+        long = "linux-agent-cache-ttl",
+        value_parser = parse_duration_secs,
+        default_value = "120"
+    )]
+    pub linux_agent_cache_ttl: Duration,
 
     /// How verbose to log
     #[arg(
@@ -398,6 +405,26 @@ mod tests {
         }
     }
 
+    #[test]
+    fn cache_ttls_have_expected_defaults() {
+        let args = parse(&[]).expect("minimal args should parse");
+        assert_eq!(args.kubelet_stats_cache_ttl, Duration::from_secs(120));
+        assert_eq!(args.linux_agent_cache_ttl, Duration::from_secs(120));
+    }
+
+    #[test]
+    fn cache_ttls_parse_from_flags() {
+        let args = parse(&[
+            "--kubelet-stats-cache-ttl",
+            "30",
+            "--linux-agent-cache-ttl",
+            "60",
+        ])
+        .expect("cache TTL flags should parse");
+        assert_eq!(args.kubelet_stats_cache_ttl, Duration::from_secs(30));
+        assert_eq!(args.linux_agent_cache_ttl, Duration::from_secs(60));
+    }
+
     #[test]
     fn push_interval_parse_correctly() {
         for interval in ["1", "80"] {

metrics-cache/src/state.rs

@@ -47,9 +47,11 @@ impl AppState<Client> {
             reader_allowlist: args.reader_allowlist.clone(),
             writer_allowlist: args.writer_allowlist.clone(),
             kubelet_stats_summary_cache: Cache::builder()
+                .time_to_live(args.kubelet_stats_cache_ttl)
                 .max_capacity(MAX_SUPPORTED_KUBERNETES_NODES)
                 .build(),
             system_agent_cache: Cache::builder()
+                .time_to_live(args.linux_agent_cache_ttl)
                 .max_capacity(MAX_SUPPORTED_KUBERNETES_NODES)
                 .build(),
             host_settings: host_settings.into(),